基于近红外高光谱成像技术的干制红枣品种鉴别

为实现干制红枣的快速鉴别，提出了一种基于近红外高光谱成像技术的鉴别方法。采集四个品种共240个样本干制红枣的近红外高光谱图像(1 000～1 600 nm)。通过主成分分析法(principal component analysis，PCA)、载荷系数法(x-Loading Weights，x-LW)和连续投影算法(successive projections algorithm，SPA)分别提取7个、8个和10个特征波长;基于灰度共生矩阵(gray level co-occurrence matrix, GLCM)提取第一主成分图像的纹理特征。分别以光谱特征、纹理特征、光谱和纹理融合特征作为输入，建立偏最小二乘判别分析(partial least squares-discriminant analysis，PLS-DA)、反向传播神经网络(back-propagation neural network，BPNN)和最小二乘支持向量机(least squares support vector machines，LS-SVM)模型。结果显示，基于融合特征的模型鉴别率高于分别基于光谱特征或纹理特征的模型鉴别率;基于融合特征的BPNN模型的结果最优，对预测集样本鉴别正确率为100%。说明近红外高光谱成像技术可用于干制红枣品种的快速鉴别。

Identification of Varieties of Dried Red Jujubes with Near-Infrared Hyperspectral Imaging

In order to realize rapid identification of dried red jujubes,this paper proposes a method based on near-infrared hyperspectral imaging technology .The near-infrared hyperspectral images (1 000～1 600 nm) of 240 samples in total from 4 cultivars of dried red jujubes will be acquired .The samples are to be divided into the calibration set and the prediction set in the ratio of 2∶1 .7,8,10 effective wavelengths are to be selected by principal component analysis(PCA),x-loading weight(x-LW)and successive projection algorithm(SPA) respectively .The dimensionality of original hyperspectral images will be reduced with PCA,and texture features of the first principal component image are to be extracted with gray-level co-occurrence matrix(GLCM) .The partial least squares-discriminant an alysis(PLS-DA),back propagation neural network(BPNN)and least square support vector machine(LS-SVM) are to be applied to build identification models with the selected effective wavelengths,texture features and fusion of the former two features .The identification rates of the models based on fusion features will b e higher than those of models based on the spectral features or texture features respectively .The BPNN models based on the fusion features will obtain the best results,whose identification rates of prediction set are to be 100％ .The results in this paper indicate that the near-infrared hyperspectral imaging techno logy has great potential to identify the dried red jujubes rapidly .